A large model-based cold storage central air conditioner time-sharing optimization control method

Abstract

The application relates to the technical field of environment control, and discloses a cold storage central air conditioner time-sharing optimization control method based on a large model, which comprises the following steps: firstly, the system cooling demand and the environment working condition prediction value at a future moment are obtained through a pre-trained deep learning model, and the current cold storage charging and discharging state is estimated in combination with a thermal front degeneration correction algorithm; subsequently, a nonlinear flow resistance model related to porosity and the cold storage state is established, and a hydraulic high energy consumption interval in which fluid resistance is highly sensitive to flow change is identified; on this basis, the energy efficiency trend of a water chilling unit is deduced by using the environment temperature to determine load distribution, and the cold storage output in the high flow resistance interval is actively inhibited through energy efficiency gate logic, so that a time sequence control strategy containing flow and load instructions is finally generated, and optimal scheduling of system-level energy efficiency is realized.

Description

Technical Field

[0001] This invention relates to the field of environmental control technology, and more specifically, to a time-sharing optimization control method for a cold storage central air conditioning system based on a large model. Background Technology

[0002] Central air conditioning systems are a major component of energy consumption in modern large buildings. To achieve peak shaving and valley filling of power grid loads and reduce operating costs, latent heat storage technology based on phase change materials (PCMs) has been widely applied. In particular, packed bed cold storage devices, which encapsulate inorganic phase change materials such as eutectic salts in spherical capsules and stack them, are widely used due to their high cold storage density and large heat exchange area. These systems typically utilize off-peak electricity pricing at night to operate chiller units and store the cooling capacity in the packed bed. During peak electricity pricing periods in the daytime, the cooling capacity is released to meet the end-user load demand, thus achieving economical operation through time-of-use pricing differences.

[0003] Existing time-of-use control methods for chilled air storage systems mostly employ optimization strategies aimed at minimizing operating costs. The conventional approach involves collecting historical load data, using linear regression or shallow neural networks to predict the total cooling load for the following day, and then planning the start-up and shutdown times of the chilled air storage tanks and the average cooling rate based on peak-valley-flat electricity pricing periods. At the hydraulic distribution and load allocation level, existing control systems typically treat the packed bed as a black box container with constant or linearly varying resistance characteristics. The control logic tends to prioritize and maximize the use of the chilled air storage side for cooling during peak electricity pricing periods. Pump speeds are usually only subject to simple PID regulation or constant differential pressure control based on end-load demand to ensure that current cooling needs are met.

[0004] However, in practical engineering applications, the aforementioned control strategies have significant energy efficiency blind spots. Eutectic salt phase change materials undergo significant volume expansion or contraction during solid-liquid transformation, leading to a nonlinear dynamic evolution of the porosity between particles within the packed bed as the cold storage state changes. Existing technologies ignore this characteristic and fail to detect the phenomenon of a sharp increase in fluid resistance during specific cold storage stages. If the control system still forcibly maintains a high flow rate for cooling according to conventional strategies at this time, the head and energy consumption of the distribution pumps will increase exponentially, easily inducing the high flow rate, low temperature difference syndrome in the chilled water system, resulting in a severe decline in the system's overall energy efficiency ratio (COP). Furthermore, existing methods lack consideration for the loss of cold energy grade (exergy) caused by the degradation of the thermal front within the packed bed, often misjudging low-grade cold energy as effective cold energy for scheduling, further exacerbating the deviation between the actual operating energy efficiency and design expectations. Summary of the Invention

[0005] This invention provides a time-sharing optimization control method for cold storage central air conditioning based on a large model, which solves the technical problems mentioned in the background art.

[0006] This invention provides a time-sharing optimization control method for a cold storage central air conditioning system based on a large model, comprising:

[0007] Historical operating parameters and future environmental conditions of the chilled water system are collected and input into a pre-trained multivariate time series prediction model, which outputs the system's cooling demand trajectory and outdoor wet-bulb temperature change trajectory during the future prediction period.

[0008] Based on the volumetric deformation characteristics of eutectic salt encapsulated capsules during phase change, a variable porosity flow resistance model is established to correlate the porosity of the packed bed with the cold storage and charging / discharging state. Based on the variable porosity flow resistance model, the flow resistance sensitivity coefficient, which characterizes the sensitivity of fluid resistance to flow rate changes, is calculated.

[0009] Based on the outdoor wet-bulb temperature change trajectory, the energy efficiency ratio change trend of the chiller unit at different times is deduced to determine the cooling weight, and the cooling load allocation value is calculated in combination with the system cooling demand trajectory;

[0010] An energy efficiency gating function is constructed using the flow resistance sensitivity coefficient to perform nonlinear correction on the cold storage load allocation value as the cold storage charging and discharging state changes. This generates a target flow setting trajectory where the flow rate decreases in the middle section and increases at both ends during the cold storage charging and discharging state. Based on this, the operating frequency of the distribution pump and the cooling load of the chiller are adjusted.

[0011] The beneficial effects of this invention include: through the deep integration of mechanism and artificial intelligence, the control system automatically avoids the high-energy-consumption hydraulic bottleneck area and naturally evolves into a dual-extreme operating trajectory with large flow at both ends and small flow in the middle section; thereby not only effectively preventing the nonlinear growth of power consumption in the transmission and distribution link and the occurrence of low temperature difference syndrome in the chilled water system, but also ensuring that the limited cooling resources are concentrated and allocated to the period when the chiller unit has the lowest operating efficiency, thus significantly improving the overall energy efficiency ratio of the central air conditioning system throughout the year. Attached Figure Description

[0012] Figure 1 This is an architecture diagram of a time-sharing optimization control system for a cold storage central air conditioning system based on a large model, according to the present invention.

[0013] Figure 2 This is a flowchart of a time-sharing optimization control method for a cold storage central air conditioning system based on a large model, according to the present invention. Detailed Implementation

[0014] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0015] like Figure 1 As shown, a time-sharing optimization control method for a cold storage central air conditioning system based on a large model includes:

[0016] Historical operating parameters and future environmental conditions of the chilled water system are collected and input into a pre-trained multivariate time series prediction model, which outputs the system's cooling demand trajectory and outdoor wet-bulb temperature change trajectory during the future prediction period.

[0017] Based on the volumetric deformation characteristics of eutectic salt encapsulated capsules during phase change, a variable porosity flow resistance model is established to correlate the porosity of the packed bed with the cold storage and charging / discharging state. Based on the variable porosity flow resistance model, the flow resistance sensitivity coefficient, which characterizes the sensitivity of fluid resistance to flow rate changes, is calculated.

[0018] Based on the outdoor wet-bulb temperature change trajectory, the energy efficiency ratio change trend of the chiller unit at different times is deduced to determine the cooling weight, and the cooling load allocation value is calculated in combination with the system cooling demand trajectory;

[0019] An energy efficiency gating function is constructed using the flow resistance sensitivity coefficient to perform nonlinear correction on the cold storage load allocation value as the cold storage charging and discharging state changes. This generates a target flow setting trajectory where the flow rate decreases in the middle section and increases at both ends during the cold storage charging and discharging state. Based on this, the operating frequency of the distribution pump and the cooling load of the chiller are adjusted.

[0020] In a preferred embodiment, the historical operating parameters and future environmental condition parameters of the chilled water system include:

[0021] The historical operating parameters and future environmental conditions of the chilled water system are constructed as an input feature vector. It is composed of static structural parameters Historically observable parameters only and known future environmental operating parameters composition:

[0022] ;

[0023] in: Including: Capsule particle size of eutectic salt encapsulated capsules Length of the filling bed Cross-sectional area of ​​the packed bed Initial porosity of the packed bed ; Including: the past Historical water supply temperature at a given moment Historical return water temperature Historical system traffic Historical packed bed pressure difference Historical water pump power Historical chiller unit power Historical cold storage charge / discharge status estimates ; Includes: covering history and the future Time-coded information at each moment, and predicted outdoor wet-bulb temperature at future moments. .

[0024] Based on the system control's demand for refined input information, parameters of different natures have different impact mechanisms on control decisions. They need to be categorized and processed according to their characteristics to improve the effectiveness of model input. Static structural parameters are inherent properties of the cold storage system, determining the basic flow resistance characteristics and heat transfer conditions of the packed bed. Only historically observable parameters record the system's past operating states, including key information such as load changes, energy consumption levels, and flow resistance evolution. This time-series data can provide training basis for the model, helping it learn the inherent laws of system operation. Among the parameters of known future environmental conditions, time-coded information can capture the daily and weekly cycle changes of air conditioning load, and the predicted outdoor wet-bulb temperature is directly related to the future operating efficiency of the chiller unit. The introduction of these two types of parameters allows the model to perceive future changes in operating conditions in advance, avoiding prediction lag caused by relying solely on historical data. Aligning these three types of parameters along the time dimension and concatenating them into an input feature vector allows the multivariate time-series prediction model to simultaneously acquire the system's physical boundaries, historical dynamics, and future environmental information, comprehensively covering key factors affecting cooling demand and environmental conditions, and providing complete data support for accurate prediction.

[0025] In a preferred embodiment, the pre-training process of the multivariate time series prediction model includes:

[0026] A training sample set is constructed using a sliding time window, and a loss function is defined. The quantile loss function is defined, and the loss function is minimized. Update the model parameters of the multivariate time series prediction model :

[0027] ;

[0028] in: These are the model parameters for a multivariate time series prediction model; A set of time indices for the training sample set; This represents the total number of prediction steps for the future prediction period; The preset set of quantiles includes 0.1, 0.5, and 0.9; For the set of quantiles Any quantile in; For the first After the moment The system requires the actual cooling capacity. For multivariate time series prediction models in the first The output at time is for the first Step and corresponding to quantile The prediction system requires cooling trajectory.

[0029] Given the temporal correlation and uncertainty of system cooling demand, traditional prediction methods struggle to capture temporal patterns under complex multivariate interactions. Therefore, a multivariate time-series prediction model is employed. A sliding time window is used to construct the training sample set to fully exploit temporal correlations in historical data. Each sample contains an input feature vector and the corresponding future system cooling demand value, enabling the model to learn the temporal mapping relationship between input parameters and future cooling demand, avoiding insufficient model generalization due to isolated data points. The quantile loss function is chosen as the training objective because system cooling demand is significantly uncertain due to random factors such as personnel movement and equipment operation. Quantile loss allows the model to output predictions at different quantiles, obtaining the most representative median prediction value while quantifying the confidence interval of the prediction. Updating the model parameters with the objective of minimizing the total loss of all training samples across all prediction steps and quantiles ensures stable prediction accuracy throughout the entire prediction time domain, avoiding excessive prediction bias in certain periods due to local optima, and ensuring that the prediction results support optimized scheduling across all time periods.

[0030] In a preferred embodiment, the trajectory of system cooling demand and the trajectory of outdoor wet-bulb temperature changes over a future forecast period include:

[0031] The system's cooling demand trajectory Trajectory of outdoor wet-bulb temperature change Selected from the median output of a multivariate time series prediction model:

[0032] ;

[0033] At the same time, calculate the current cold storage charge / discharge status. This includes calculating the thermal front degradation factor. Effective cold energy storage and the state of cold storage charging and discharging :

[0034] ;

[0035] ;

[0036] ;

[0037] in: This represents the model output corresponding to the 0.5 quantile; The sensitivity coefficient of the thermal front; for The temperature difference between the supply and return water at any given time; To control the time step; For fluid density; Specific heat capacity of the fluid; for System traffic at any given moment; , They are respectively The return water temperature and supply water temperature at any given time; For reference energy standard; To maximize cold storage capacity; This is the Sigmoid smoothing function.

[0038] To ensure the reliability of the prediction results and the accurate quantification of the cold storage status, the median probability prediction value was chosen as the trajectory of the system's cooling demand and the outdoor wet-bulb temperature change. This is because the median reflects the central tendency of the data and is more robust than extreme quantiles, avoiding overly aggressive or conservative predictions. When calculating the cold storage charging and discharging status, the thermal stratification inside the packed bed gradually degrades during operation. This degradation leads to a decrease in the quality of the cooling capacity. Ignoring this loss would result in low-quality cooling capacity being misclassified as effective cooling capacity, thus affecting the accuracy of scheduling decisions. Therefore, a thermal front degradation factor is introduced. The degree of thermal stratification degradation is quantified by the rate of change of the supply and return water temperature difference, thereby correcting the instantaneous cold storage power. Integrating the corrected instantaneous cold storage power over time is to accumulate the actual usable effective cold storage energy. Normalizing the effective cold storage energy to between 0 and 1 using a smoothing mapping function is to establish a unified quantitative standard for cold storage state. At the same time, the smoothing mapping can avoid numerical out-of-bounds occurrences of state values ​​at physical boundaries, accurately reflect the actual characteristics of the nonlinear decay of the heat absorption and release capacity of eutectic salt at the end of the phase transition, and ensure that the calculation of the cold storage charge and release state conforms to the law of system operation.

[0039] In a preferred embodiment, based on the volumetric deformation characteristics of the eutectic salt encapsulation capsule during the phase transition process, a variable porosity flow resistance model is established that correlates the porosity of the packed bed with the cold storage and discharging state, including:

[0040] Define the porosity of a packed bed With cold storage charge and discharge state The relationship between the changes was analyzed, and a variable porosity flow resistance model was constructed based on the Eugen equation to calculate the pressure difference in the packed bed. :

[0041] ;

[0042] ;

[0043] in: The initial porosity of the packed bed; The mid-section indentation strength coefficient is greater than zero; It is in a cold storage charging and discharging state; For fluid dynamic viscosity; The length of the filling bed; The particle size of the eutectic salt encapsulated capsules; For fluid density; The apparent velocity of the fluid within the packed bed is equal to the system flow rate. Divide by the cross-sectional area of ​​the packed bed .

[0044] Current technologies treat packed beds as fixed-resistance containers, failing to address porosity changes caused by phase transitions. Therefore, a variable-porosity flow resistance model is needed. Eutectic salts exhibit the most significant volume expansion during the mid-phase transition, compressing the internal voids of the packed bed and leading to a significant reduction in porosity. The phase transition at the two ends is weaker, with porosity approaching the initial state. This variation closely matches the concave morphology of a quadratic function; therefore, the packed bed porosity is defined as a quadratic function of the cold storage and dissipation state. The mid-phase concave strength coefficient is used to accurately match the volume expansion amplitude of different eutectic salt materials. The Eugen equation is a mature method for describing the flow resistance of porous media, encompassing both viscous and inertial drag. By substituting the dynamically changing packed bed porosity and the apparent fluid velocity, a nonlinear model reflecting the true flow resistance characteristics can be constructed. Fluid dynamic viscosity and fluid density are the fundamental thermophysical parameters for flow resistance calculation. The apparent fluid velocity is converted from the system flow rate to the packed bed cross-sectional area to ensure consistency between the model input and the actual flow state. The packed bed pressure difference serves as a key indicator for model verification, ensuring the accuracy of the model calculations.

[0045] In a preferred embodiment, the flow resistance sensitivity coefficient, characterizing the sensitivity of fluid resistance to flow rate changes, is calculated based on a variable porosity flow resistance model, including:

[0046] Calculate the flow resistance sensitivity coefficient The calculation formula is as follows:

[0047] ;

[0048] in: This is the flow resistance sensitivity coefficient; To follow the cold storage charge and discharge state The formula characterizes the change in packed bed porosity; When the fluid resistance decreases in the middle of the cold storage charging and discharging state, it exhibits a non-linear growth trend on the order of inverse cubic.

[0049] Based on the extraction of core influencing factors from the variable porosity flow resistance model, the key cause of flow resistance variation is the nonlinear evolution of porosity, and the amplification effect of this evolution on system energy consumption needs to be quantified. In the Eugen equation, the influence of porosity on resistance is reflected in a specific combination term: the ratio of 1 minus the difference in current packed bed porosity to the cube of the current packed bed porosity. This effectively amplifies the resistance fluctuations caused by small changes in porosity, especially when the porosity is low in the middle section, where this ratio increases sharply, forming a significant peak, which corresponds precisely to the high hydraulic resistance range. By constructing a flow resistance sensitivity coefficient through this combination, no additional measuring equipment is required; it can be calculated solely from the packed bed porosity in the variable porosity flow resistance model, enabling the control system to quickly identify high-energy-consuming hydraulic ranges.

[0050] In a preferred embodiment, the energy efficiency ratio (EER) of the chiller unit at different times is deduced based on the outdoor wet-bulb temperature change trajectory to determine the cooling load weight, and the cooling load allocation value is calculated in conjunction with the system cooling demand trajectory, including:

[0051] First, predict the future... Chiller unit energy efficiency ratio at any time :

[0052] ;

[0053] Next, calculate the future number. The weight of cooling down at any moment :

[0054] ;

[0055] Finally, calculate the future number. Cooling load distribution value at any time :

[0056] ;

[0057] in: These are the fitting coefficients for the energy efficiency model of the chiller unit; The outdoor wet-bulb temperature change trajectory in The value at time; The system's cooling load trajectory is in The value at time; This is the weighting sharpness parameter, used to adjust the degree of tilt towards periods of low energy efficiency; To predict the total number of steps; To maximize cold storage capacity; This represents the current state of cold storage charging and discharging. To ensure the minimum cold storage charge and discharge state; The slope parameter of the Sigmoid function; For the Sigmoid function; To control the time step.

[0058] The energy efficiency ratio (EER) of chillers is significantly affected by the outdoor wet-bulb temperature and the system cooling demand. The outdoor wet-bulb temperature determines the cooling tower's heat dissipation efficiency, while the system cooling demand affects the unit's load rate. Therefore, a linear regression model combining these two parameters is used to predict the future EER. The fitting coefficients of the chiller's energy efficiency model are obtained by fitting historical operating data to ensure prediction accuracy. To concentrate limited cold storage resources on the chiller's inefficient periods, an exponential function relating the reciprocal of the chiller's EER is constructed. The lower the EER, the larger the reciprocal, and the higher the corresponding cooling weight. The weight sharpness parameter is used to adjust the concentration of weights, avoiding overly dispersed or extreme allocation. Normalization ensures that the sum of the weights at all times is 1, satisfying the total allocation constraint. The calculation of total available cooling capacity needs to take into account both the current cooling storage charging and discharging status and the backup cooling storage status. The backup cooling storage status is used to reserve emergency cooling capacity. The slope parameter of the Sigmoid function adjusts the sensitivity of the cooling intention to changes with the cooling storage status, so as to avoid releasing a large amount of cooling when the cooling storage capacity is too low. Finally, the total available cooling capacity is allocated according to the cooling release weight to obtain a cooling storage load allocation value sequence that meets both the terminal demand and the energy efficiency optimization.

[0059] In a preferred embodiment, the energy efficiency gating function is constructed using the flow resistance sensitivity coefficient, including:

[0060] Constructing energy efficiency gating functions The calculation formula is as follows:

[0061] ;

[0062] in: It is an energy efficiency gating function; This is the pinch penalty strength parameter, used to set the suppression strength of the hydraulic pinch area; This is the flow resistance sensitivity coefficient; This represents the current state of cold storage charging and discharging.

[0063] To suppress cold storage output in high flow resistance ranges, the flow resistance sensitivity coefficient can accurately identify high-energy-consuming hydraulic regions. A specific function is needed to correlate this coefficient with the cold storage output weight to achieve dynamic control. The energy efficiency gating function is negatively correlated with the flow resistance sensitivity coefficient; the higher the flow resistance sensitivity coefficient, the lower the cold storage output weight should be to avoid a surge in energy consumption caused by high flow rate operation under high flow resistance. The denominator is designed as the sum of a numerical value, the pinch penalty strength parameter, and the flow resistance sensitivity coefficient. This allows for flexible adjustment of the suppression level through the pinch penalty strength parameter to adapt to the flow resistance characteristics of different systems, while ensuring the function output remains within a reasonable range. Precise suppression of high flow resistance ranges can be achieved using only existing parameters, allowing the control system to autonomously adapt to changes in the cold storage charging and discharging state.

[0064] In a preferred embodiment, the cold storage load allocation value is nonlinearly corrected according to the cold storage charging and discharging state, generating a target flow setting trajectory in which the flow rate decreases in the middle section and increases at both ends during the cold storage charging and discharging state, including:

[0065] First, calculate the corrected future number. Target cold storage capacity at any time :

[0066] ;

[0067] Subsequently, the future number of... is calculated based on the low temperature difference risk constraint. Target flow setting trajectory at any time :

[0068] ;

[0069] in: For the future The energy efficiency gating function value corresponding to the given time; For the future Predicted values ​​of the cold storage charge / discharge state at any given time; Values ​​allocated to cold storage load; For smooth activation functions, it is defined as follows: ; For fluid density; Specific heat capacity of the fluid; Design the supply and return water temperature difference for the chilled water system; This is the low temperature difference sensitivity coefficient, used to characterize the degree to which an increase in flow rate leads to a decrease in the temperature difference between the supply and return water. This is the smoothing scale parameter.

[0070] The chilled water load allocation value is derived from chiller efficiency optimization, but it does not consider flow resistance characteristics and needs to be corrected by combining it with an energy efficiency gating function. By multiplying the energy efficiency gating function with the chilled water load allocation value, the target chilled water power after hydraulic pinch suppression can be obtained, ensuring power reduction in the middle stage of chilled water charging and discharging. Considering that chilled water systems are prone to the syndrome of large flow rate and small temperature difference, a flow coupling model needs to be constructed to limit excessive flow rate increase. The supply and return water temperature difference is designed as a benchmark, and the low temperature difference sensitivity coefficient quantifies the impact of increased flow rate on temperature difference attenuation. The combination of the two forms the flow constraint boundary. The introduction of a smoothing activation function is to avoid abrupt changes in flow rate calculation. The smoothing scale parameter adjusts the curvature of the function to ensure that the flow rate changes smoothly with the target chilled water power. The final target flow rate setting trajectory satisfies the cooling demand while avoiding the risk of energy efficiency degradation.

[0071] In a preferred embodiment, adjusting the operating frequency of the water distribution pump and the cooling load of the chiller unit includes:

[0072] Calculate the future Chiller cooling load at all times The calculation formula is as follows:

[0073] ;

[0074] Calculate the future Water pump operating frequency set value at any time The calculation formula is as follows:

[0075] ;

[0076] in: For the system's cooling load trajectory in The predicted value at any given time; The target cold storage capacity after energy efficiency gating correction; This refers to the rated operating frequency of the water distribution pump. Set a trajectory for the target traffic The value at time; This is the rated flow rate of the chilled water system.

[0077] Based on system cooling capacity balance and equipment operating characteristics, the executability of control commands is ensured. To ensure that the cooling capacity demand of the terminal load is not met, the target chilled water storage power sequence is subtracted from the system's required cooling capacity trajectory to obtain the chiller unit's cooling load. When the target chilled water storage power decreases during the charging and discharging phase, the chiller unit automatically takes on more load, forming a complementary relationship. The operating frequency adjustment of the distribution pumps must follow their own operating rules. The pump similarity law clearly states the direct proportionality between flow rate and frequency. Therefore, based on the ratio of the target flow rate setpoint trajectory to the rated flow rate of the chilled water system, combined with the rated operating frequency of the distribution pumps, the pump operating frequency setpoint at each moment can be calculated. Issuing the chiller unit's cooling load and the pump operating frequency setpoint as timing control commands ensures that the underlying equipment accurately executes the optimization strategy, achieving system-level collaborative control.

[0078] The capsule size of eutectic salt encapsulation capsules refers to the diameter of the spherical capsules encapsulating the eutectic salt phase change material. It is a key structural parameter for the flow resistance characteristics of packed beds and directly affects the weights of the viscous and inertial terms in the Eugen equation. This value can be obtained by consulting the equipment nameplate or project as-built drawings. The preferred value range is 3 cm to 10 cm, with 5 cm being commonly chosen in typical applications.

[0079] The bed length of a packed bed refers to the total length of the packed bed formed by the stacking of eutectic salt encapsulated capsules in the direction of fluid flow, and is a fundamental parameter for calculating fluid flow resistance. It is obtained by measuring the actual length of the packed area inside the cold storage tank, with a preferred value ranging from 2 meters to 5 meters, and a typical value of 3.5 meters.

[0080] The cross-sectional area of ​​a packed bed refers to the cross-sectional area of ​​the packed bed perpendicular to the direction of fluid flow, used to calculate the apparent velocity of the fluid within the packed bed. It is calculated based on the inner diameter of the cold storage tank; for example, for a circular cold storage tank, the area can be calculated using the formula for the area of ​​a circle after measuring the inner diameter, with a typical value of 12.5 square meters.

[0081] The initial porosity of a packed bed refers to the ratio of the void volume between capsule particles to the total volume of the packed bed under fully cooled or fully vented conditions. It is a benchmark parameter for flow resistance model calculations. Determined by the capsule packing method, the preferred value range for randomly packed beds is 0.35 to 0.45, with a typical value of 0.4.

[0082] Historical supply water temperature refers to the chilled water temperature data output from the chiller unit or cold storage tank over a past period, used to reflect historical heat load levels and model training. It is collected from the system's historical database and must maintain consistency with the time dimension of other parameters, with a sampling step of 15 minutes.

[0083] Historical return water temperature refers to the temperature data of chilled water returning from the terminal load to the chiller unit or cold storage tank over a past period. Combined with historical supply water temperature, the historical supply and return water temperature difference can be calculated. Data is collected from the system's historical database, with a sampling step size consistent with the historical supply water temperature. Standardization processing is required before inputting this data into the model.

[0084] Historical system flow rate refers to the total circulating water flow of the chilled water system over a past period, and is an important parameter for calculating historical cooling capacity and flow resistance characteristics. It is obtained through historical data from flow meters, measured in cubic meters per second, and needs to be collected synchronously with supply and return water temperature data.

[0085] Historical packed bed differential pressure refers to the pressure difference between the inlet and outlet of the packed bed over a past period, implicitly containing information about historical flow resistance characteristics. It is obtained through historical monitoring data from differential pressure sensors, measured in Pascals, and used to assist models in learning the relationship between flow resistance, flow rate, and porosity.

[0086] Historical pump power refers to the power consumed by the distribution pumps during past operation, and is a key data point reflecting historical hydraulic energy consumption. Extracted from pump operation records or power monitoring systems, it is measured in kilowatts and used as an energy consumption benchmark for model training.

[0087] Historical chiller unit power refers to the power consumed by the chiller unit during past cooling operations, reflecting historical cooling energy consumption levels. It is obtained through the chiller unit's operating logs or electricity metering system, and is measured in kilowatts. This power consumption is used in conjunction with historical cooling load data for model training.

[0088] Historical cold storage charge / discharge state estimates refer to the cold storage state values ​​calculated and normalized from the effective cold storage energy at past moments. The values ​​range from 0 to 1, where 0 represents complete cold release and 1 represents complete cold storage charge. These are not directly measured values, but are calculated using historical supply and return water temperature differences, flow rates, and other data, and are used as state characteristics for model input.

[0089] Time-coded information refers to hourly and weekly periodic feature data covering historical and future forecast periods, used to allow models to learn the daily and weekly cycle patterns of air conditioning load. Hours (0 to 23) and weeks (0 to 6) are mapped to continuous vectors using sine and cosine functions. For example, hourly codes are sine (2π multiplied by the hour number divided by 24) and cosine (2π multiplied by the hour number divided by 24), and weekday codes are encoded similarly.

[0090] The predicted outdoor wet-bulb temperature refers to the outdoor wet-bulb temperature data at every moment within the predicted period. It is a core environmental parameter for extrapolating changes in the energy efficiency ratio of chillers. Data is obtained from a meteorological service interface, updated at least once per hour, and interpolated to a 15-minute sampling step to maintain consistency with the time granularity of the system operating parameters.

[0091] The input feature vector is a complete data vector formed by concatenating static structural parameters, historically observable parameters, and known future environmental parameters along the time dimension. It is used as input to a multivariate time series prediction model. The vector contains the system's physical boundary conditions, historical operational dynamics, and future environmental information, ensuring that the model can comprehensively perceive influencing factors.

[0092] The historical lookback window length refers to the number of historical time steps traced back during model training and prediction, used to construct the model's input historical sequence. Based on the thermal inertia of the central air conditioning system and the periodicity of time-of-use electricity prices, a preferred value is 96 steps, corresponding to the past 24 hours, with each step having a sampling time step of 15 minutes.

[0093] The future prediction time domain length refers to the number of future time steps the model needs to output the prediction results to cover the complete optimization scheduling cycle. The preferred value is 96 steps, corresponding to the next 24 hours, with each step having a sampling time step of 15 minutes, consistent with the time granularity of the historical review window.

[0094] The control time step refers to the time interval between system data acquisition, model prediction updates, and control command issuance; it is the basic time unit for timing control. A preferred value is 15 minutes (900 seconds), balancing control accuracy and system computational efficiency, and aligning with the time-of-use pricing in the electricity market.

[0095] The model parameters of a multivariate time series prediction model refer to all adjustable weights and biases in the model (such as the time fusion Transformer model), which determine the model's predictive ability. Updates are performed by minimizing the quantile loss function. During training, the Adam optimizer is used, with a preferred learning rate of 0.001 and a preferred batch size of 64. Training is iteratively continued until the loss on the validation set no longer decreases, at which point the model is fixed.

[0096] The time index set of the training sample set refers to the set of indices for all historical time points used for model training, with each index corresponding to a complete training sample. It is extracted from the historical database using a sliding time window, with a window sliding step of one step, ensuring full utilization of historical data while preserving temporal relevance.

[0097] The preset quantile set refers to the set of quantile values ​​set during model training and prediction, used to quantify the uncertainty of the prediction results. It is fixed to include three quantiles: 0.1, 0.5, and 0.9. Among them, 0.5 corresponds to the median prediction value and is used for deterministic control input; 0.1 and 0.9 correspond to the 10% and 90% quantiles, respectively, and are used to construct the prediction confidence interval.

[0098] Any quantile in the quantile set refers to a single value in the preset quantile set, used to calculate the prediction loss at that quantile. During training, the regression loss between the predicted value and the true value at each quantile needs to be calculated separately, and then the losses of all quantiles are summed as the total loss of the model.

[0099] The true cooling load requirement of the system at step h after time t refers to the actual cooling load requirement of the system at the h-th control time step after time t in the historical time. It is the label data for model training. It is obtained through cooling load measurement data in the historical operation of the system to ensure a one-to-one correspondence with the time step of the model input feature vector.

[0100] The predicted system cooling demand trajectory output by the multivariate time-series forecasting model at time t, for the h-th step and corresponding to quantile τ, refers to the predicted system cooling demand value output by the model at time t in history, for the h-th future control time step, corresponding to quantile τ. This predicted value is used in conjunction with the actual value to calculate the quantile regression loss, which is used for model parameter updates. The predicted values ​​at different quantiles together constitute the probability distribution of future cooling demand.

[0101] The quantile loss function is the objective function used in model training to measure the deviation between predicted and true values ​​at different quantiles, effectively quantifying the uncertainty of prediction. By calculating the sum of losses for each training sample at all prediction steps and all quantiles, the model parameters are updated with the goal of minimizing this sum, ensuring that the model can accurately output prediction results at different quantiles.

[0102] The learning rate refers to the step size of parameter updates during model training, used to control the model's convergence speed and training stability. A value of 0.001 is preferred to avoid training oscillations caused by an excessively large learning rate or low training efficiency caused by an excessively small learning rate.

[0103] Batch size refers to the number of training samples input each time the model is trained, used to balance training efficiency and memory usage. A preferred value is 64, which ensures both the stability of the training process and makes full use of computing resources to improve training speed.

[0104] The system cooling demand trajectory for the future forecast period refers to the sequence of system cooling demand for each control time step within the next 24 hours predicted by the model. It serves as the basis for load allocation and optimized control. The 0.5 quantile (median) predicted value of the model output is selected to ensure the robustness of the forecast results and avoid overly aggressive or conservative control strategies caused by extreme quantiles.

[0105] The outdoor wet-bulb temperature trajectory within the future forecast period refers to the outdoor wet-bulb temperature sequence at each control time step within the next 24 hours predicted by the model, used to extrapolate the energy efficiency ratio (EER) trend of the chiller unit. It is output synchronously with the system cooling demand trajectory, also using the 0.5 quantile forecast value to ensure consistency with load forecasts.

[0106] The output of the multivariate time-series forecasting model corresponding to the 0.5 quantile refers to the prediction result of the multivariate time-series forecasting model at the 0.5 quantile, which includes two parts: the future cooling demand trajectory and the outdoor wet-bulb temperature trajectory. This output is the core input for subsequent optimization calculations, integrating the model's judgment on the most likely future operating conditions.

[0107] The current cold storage charge / discharge state refers to the normalized state value of the effective cold storage energy of the cold storage tank at the current moment, with a value ranging from 0 to 1. It is used to reflect the remaining cold capacity level of the cold storage tank. 0 represents a fully discharged state, and 1 represents a fully charged state. It is calculated from the effective cold storage energy through a smoothing mapping function and is a key state parameter for flow resistance model and load distribution.

[0108] The thermal front degradation factor is a dimensionless parameter characterizing the loss of cooling capacity due to thermal stratification degradation within the packed bed. It is used to correct instantaneous cooling capacity and prevent low-grade cooling capacity from being misclassified as effective cooling capacity. It is calculated by the rate of change of the supply and return water temperature difference between the current and previous moments. The larger the rate of change, the smaller the degradation factor, and the more severe the cooling capacity loss. Its calculation formula is based on an exponential function to ensure the rationality of the loss quantification.

[0109] The thermal front sensitivity coefficient is a dimensionless empirical parameter used to adjust the decay rate of the thermal front degradation factor, controlling the impact of temperature fluctuations on the loss of cooling capacity. It is obtained through offline experimental calibration, with an optimal value range of 0.1 to 0.5 and a typical value of 0.2. This means that when temperature fluctuations are large, the effective cooling capacity will be discounted exponentially.

[0110] The supply and return water temperature difference at time t refers to the difference between the supply water temperature and the return water temperature of the chilled water system within the current control time step, and is a core parameter for calculating instantaneous cooling capacity. It is calculated directly by collecting the supply and return water temperatures in real time, and the unit is degrees Celsius.

[0111] The supply and return water temperature difference at time t-1 refers to the difference between the supply and return water temperatures of the chilled water system within the previous control time step, and is used to calculate the rate of change of the supply and return water temperature difference. Combined with the supply and return water temperature difference at the current time, it helps determine the magnitude of the thermal front degradation factor, which is also calculated using real-time data acquisition.

[0112] The effective cold storage energy at time t refers to the total amount of cold energy actually available in the cold storage tank at that time, and is the core energy parameter for measuring the cold storage status. It is calculated by adding the effective cold storage energy at the previous time to the product of the instantaneous cold storage power at the current time and the thermal front degradation factor, and then multiplying by the control time step, with the unit being kilowatt-hours.

[0113] The cumulative effective cold storage energy at time t-1 refers to the total amount of actually usable cold energy in the cold storage tank at the end of the previous control time step, and is the basis for calculating the current effective cold storage energy. It is obtained by storing historical calculation results to ensure the continuity and accuracy of energy calculations.

[0114] Fluid density refers to the density of chilled water, used to calculate the mass flow rate and cooling capacity of chilled water. At the operating temperature of chilled water (approximately 5 degrees Celsius), a value of 998 kg / m³ is preferred and can be considered a constant for engineering calculations.

[0115] Specific heat capacity of a fluid refers to the isobaric specific heat capacity of chilled water, which is the amount of heat required to raise the temperature of a unit mass of chilled water by 1 degree Celsius. It is a key thermophysical parameter for calculating cooling capacity. A preferred value is 4.18 kJ / (kg·Kelvin), which is suitable for common concentrations of ethylene glycol chilled water or pure water chilled water.

[0116] The system flow rate at time t refers to the circulating flow rate of the chilled water system within the current control time step, used to calculate the instantaneous cooling capacity. It is acquired through a real-time flow meter, with units of cubic meters per second, and needs to be collected synchronously with the supply and return water temperature difference data to ensure accurate cooling capacity calculation.

[0117] The return water temperature at time t refers to the temperature of the chilled water returning from the terminal load within the current control time step. It is collected in real time by a temperature sensor installed in the return water pipeline and is measured in degrees Celsius. It is the basic data for calculating the supply and return water temperature difference and cooling capacity.

[0118] The supply water temperature at time t refers to the temperature of the chilled water output from the cold storage tank or chiller unit within the current control time step. It is collected in real time by a temperature sensor installed in the supply water pipeline and is measured in degrees Celsius. It is used in conjunction with the return water temperature.

[0119] The reference energy baseline refers to the effective cold storage energy level corresponding to a cold storage charge / discharge state of 0.5. It is a baseline parameter that maps the effective cold storage energy to a normalized cold storage state. It is usually taken as half of the total cold storage capacity designed for the cold storage tank to ensure a symmetrical distribution of the cold storage state between 0 and 1.

[0120] Effective cold storage capacity refers to the characteristic scaling factor used to adjust the slope of the cold storage charge-discharge state mapping curve, ensuring a smooth transition of the cold storage state between 0 and 1 with saturation characteristics at both ends. The preferred value is one-quarter to one-sixth of the total latent heat of phase change in the system design, avoiding numerical out-of-bounds occurrences of the cold storage state at physical boundaries.

[0121] The Sigmoid smoothing function is a nonlinear mapping function used to normalize the effective cold storage energy to the cold storage charge / discharge state, and its output range is strictly limited to 0 to 1. This function can accurately reflect the nonlinear decay of the heat absorption and release capacity of eutectic salts at the end of the phase transition, ensuring the physical rationality of the cold storage state calculation.

[0122] The porosity of a packed bed as a function of the cold storage and charge / discharge state refers to the dynamic evolution of the porosity of the packed bed under these conditions. It is primarily used to reflect the influence of the volumetric deformation of the eutectic salt encapsulation capsule on the bed structure. Its variation is described by a quadratic function, exhibiting a concave shape in the middle section of the cold storage and charge / discharge state, where the porosity is lowest, and approaching the initial porosity at both ends. This parameter does not require direct measurement; it is calculated using the initial porosity, the mid-section concave strength coefficient, and the current cold storage and charge / discharge state. Its value ranges with variations in initial porosity and the mid-section concave strength coefficient, typically fluctuating between 0.35 and 0.42.

[0123] The mid-section depression strength coefficient is a dimensionless empirical parameter used to adjust the degree of mid-section depression in the packed bed porosity during cold storage and discharging. It is used to quantify the reduction in porosity caused by the volume expansion of the eutectic salt encapsulation capsule during the phase change phase. It is obtained through offline hydraulic experiments, with an optimal value range of 0.05 to 0.15 and a typical value of 0.1. A higher value indicates a more significant reduction in mid-section porosity and a higher peak hydraulic resistance.

[0124] Dynamic viscosity refers to the viscosity coefficient of chilled water at its operating temperature, and is a key thermophysical parameter for calculating viscous resistance losses in packed beds. This parameter is related to the temperature and composition of the chilled water (such as ethylene glycol concentration). At common operating temperatures for chilled water (around 5 degrees Celsius), a preferred value range is 1.3 multiplied by 10. -3 Up to 1.6 x 10 -3 Pascals per second can be obtained by consulting a table of properties of frozen water or by experimental measurement.

[0125] The apparent velocity of fluid in a packed bed refers to the average flow velocity of chilled water across the cross-section of the packed bed. It is a key parameter connecting the system flow rate and the flow state inside the packed bed. It is calculated by dividing the system flow rate by the cross-sectional area of ​​the packed bed, and the unit is meters per second. For example, when the system flow rate is 0.1 cubic meters per second and the cross-sectional area of ​​the packed bed is 12.5 square meters, the apparent velocity is 0.008 meters per second.

[0126] The packed bed differential pressure refers to the pressure difference between the inlet and outlet of the chilled water flowing through the packed bed. It is a direct parameter reflecting the flow resistance characteristics of the packed bed and is used to verify the accuracy of the variable porosity flow resistance model. The differential pressure is collected in real time by differential pressure sensors installed at the inlet and outlet of the packed bed, and the unit is Pascal. Its value varies with porosity and apparent flow velocity, and it exhibits a high value in the middle section under cold storage conditions due to the low porosity.

[0127] The flow resistance sensitivity coefficient is a dimensionless parameter that quantifies the amplification effect of the packed bed porosity on system energy consumption under the current cold storage charging and discharging state. Its core function is to identify the high hydraulic resistance range. The calculation logic is to subtract the current packed bed porosity from one and divide by the cube of the current packed bed porosity. This parameter does not require direct measurement; it is calculated from the packed bed porosity as it changes with the cold storage charging and discharging state. Its characteristic value is that it reaches its peak in the middle of the cold storage charging and discharging state, typically ranging from 10 to 12, with values ​​at both ends of approximately 9.3 to 9.5. The cold storage charging and discharging state range corresponding to the peak value is the high hydraulic energy consumption range.

[0128] The energy efficiency ratio (EER) of a chiller unit refers to the ratio of the cooling capacity output by the chiller unit to the electrical power consumed at a future point in time. It is a core parameter for measuring the operating efficiency of a chiller unit and determines the allocation logic of cooling load. It is predicted using a linear regression model, with the outdoor wet-bulb temperature and system cooling demand at the corresponding time as inputs. The prediction result directly reflects the energy-saving potential of the chiller unit under that operating condition, typically ranging from 3.0 to 6.0. A lower value indicates lower operating efficiency of the chiller unit and a greater need for a cold storage system to supplement cooling.

[0129] The fitting coefficients of the chiller unit energy efficiency model refer to the constant parameters used in the linear regression model to fit the relationship between the chiller unit's energy efficiency ratio, outdoor wet-bulb temperature, and system cooling demand. These include the intercept term and two variable coefficients. The coefficients are obtained by fitting the model using the least squares method with historical operating data of the chiller unit (such as the energy efficiency ratio, outdoor wet-bulb temperature, and system cooling demand records for the past year). Typical values ​​are an intercept term of 5.5, an outdoor wet-bulb temperature coefficient of -0.15, and a system cooling demand coefficient of 0.002. The coefficient of determination (R²) of the fitted model is... 2 The value should be no less than 0.85 to ensure prediction accuracy.

[0130] The weighting sharpness parameter is a dimensionless parameter that adjusts the concentration of cooling load, used to control the degree to which cold storage resources are tilted towards the low-efficiency periods of the chiller unit. It is a preset parameter, with an optimal value range of 5 to 8, and a typical value of 6. A larger value concentrates the weight more precisely on the few times when the chiller unit's energy efficiency ratio is lowest, resulting in a more significant peak-shaving effect; a smaller value results in a smoother weight distribution and more stable system operation. Specific values ​​can be determined through full-cycle energy consumption simulation optimization.

[0131] The total number of prediction steps refers to the total number of control time steps included in the future prediction time domain of the multivariate time series prediction model. It is used as the time period allocation base for calculating the total available cooling capacity. The preferred value is 96 steps, corresponding to the next 24 hours, with each step having a time step of 15 minutes.

[0132] The minimum cold storage charge / discharge state refers to the lowest cold storage state value reserved to cope with unexpected situations such as sudden failures and extreme high temperatures, used to ensure the safety and reliability of system operation. This is a preset parameter, with an optimal value range of 0.1 to 0.2, and a typical value of 0.15. The value is determined through engineering experience and needs to be comprehensively set in conjunction with the building's emergency cooling needs and the cold storage tank capacity to ensure that the reserved cooling capacity can meet at least two hours of emergency cooling requirements.

[0133] The slope parameter of the Sigmoid function refers to the parameter that adjusts the inclination of the Sigmoid function curve, used to control the sensitivity of the cooling intention to changes in the current cold storage charging and discharging state. This is a preset parameter, preferably ranging from 10 to 20, with a typical value of 15. A larger value results in a faster increase in cooling intention when the current cold storage charging and discharging state is higher than the minimum cold storage charging and discharging state, and vice versa. The optimal value is determined through offline simulation to balance cooling intention and inventory protection.

[0134] Cooling release weight refers to the weight allocated to the release of stored cold energy at each moment within the predicted future period. Its core function is to concentrate stored cold energy resources on periods when chillers are less efficient. It is calculated using an exponential function of the inverse of the chiller's energy efficiency ratio and then normalized. The weight for each moment ranges from 0 to 1, and the sum of the weights for all moments is 1. The lower the chiller's energy efficiency ratio, the larger the weight value at each moment, and the more stored cold energy is allocated.

[0135] The cold storage load allocation value refers to the rated cooling power that the cold storage system should undertake within the (t+k)th control time step in the future. It is the basic command parameter for the output of the cold storage system. It is calculated by allocating the total available cold storage capacity according to the cold storage weight, and the unit is kilowatts. The total available cold storage capacity is determined by mapping the effective cold storage capacity, the difference between the current cold storage charge / discharge state and the minimum cold storage charge / discharge state through the Sigmoid function. After allocation, it ensures that the cold storage output of the chiller unit is more sufficient during the inefficient period.

[0136] The energy efficiency gating function is a nonlinear function used to reduce the weight of cold storage output in regions with high fluid resistance sensitivity. Its core function is to actively suppress the cold storage power in high flow resistance regions, preventing a surge in energy consumption of distribution pumps. It is negatively correlated with the flow resistance sensitivity coefficient. The calculation logic is to divide a numerical value by a specific denominator, where the denominator is the sum of the products of the numerical value, the pinch penalty strength parameter, and the flow resistance sensitivity coefficient. This function does not require direct measurement; it is calculated using the pinch penalty strength parameter and the flow resistance sensitivity coefficient, with a value ranging from 0.45 to 0.51. The value is lowest in the middle of the cold storage charging / discharging state and higher at both ends.

[0137] The pinch penalty intensity parameter is a dimensionless hyperparameter that adjusts the degree of suppression in the high flow resistance region. It is used to set the degree of aversion of the control system to the high energy consumption region of hydraulics. It is a preset parameter, with an optimal value range of 0.05 to 0.2, and a typical value of 0.1. A larger value results in stronger suppression of the cold storage output in the high flow resistance region. The optimal value needs to be determined through offline system simulation. The simulation aims to minimize the total energy consumption (pump consumption + unit power consumption) over the entire cycle using a grid search.

[0138] The target cold storage power refers to the actual cooling power that the cold storage system should undertake within the (t+k)th control time step after hydraulic pinch suppression, and it serves as the basis for the final cold storage output. It is obtained by multiplying the energy efficiency threshold function value and the cold storage load allocation value, and is expressed in kilowatts. During the middle of the cold storage charging and discharging process, due to the decrease in the energy efficiency threshold function value, this power is significantly lower than the cold storage load allocation value; at both ends, it is close to the cold storage load allocation value, ensuring that high-energy-consuming hydraulic zones are avoided.

[0139] The energy efficiency gating function value at time t+k in the future refers to the specific output value of the energy efficiency gating function at the control time step t+k in the future, which is used to correct the cold storage load allocation value at that time. The time-series output sequence is the same function as the energy efficiency gating function, specifically referring to the value at time t+k in the sequence. It is calculated by the predicted value of the cold storage charging and discharging state at that time in the future and the flow resistance sensitivity coefficient. The value changes with the future cold storage state, being lower in the middle and higher at both ends.

[0140] The predicted value of the cold storage charging and discharging state at time t+k refers to the estimated value of the cold storage charging and discharging state at the next control time step (t+k), ranging from 0 to 1. It is used to calculate the energy efficiency gating function value at the corresponding time. This value is derived by combining data such as the current cold storage charging and discharging state and the future cooling demand trajectory through a multivariate time-series prediction model. It needs to be matched with the future flow resistance characteristic change trend to ensure the forward-looking nature of the energy efficiency gating correction.

[0141] The target flow rate setting trajectory refers to the target circulating flow rate sequence for each control time step within the future forecast period, and is the core instruction basis for adjusting the operating frequency of the distribution pumps. It is calculated by combining the target cold storage power sequence with the flow coupling model, with the unit being cubic meters per second. Its characteristics are that the flow rate decreases in the middle section and increases at both ends during the cold storage charging and discharging state, which satisfies the cooling demand while avoiding the energy efficiency deterioration caused by large flow rates.

[0142] The smoothing activation function is a nonlinear function used to ensure smooth and differentiable flow rate calculations and avoid abrupt numerical changes. Its core function is to limit excessive flow rate increases through smoothing constraints. It is defined as a specific value multiplied by the natural logarithm, where the logarithm is the sum of the value and a specific power of the natural constant. The power of the natural constant is the ratio of the input value to the smoothing scale parameter. The input value consists of the product of fluid density, fluid specific heat capacity, and design supply and return water temperature difference, minus the product of the low temperature difference sensitivity coefficient and the target cold storage power, ensuring that the flow rate changes smoothly with the cooling demand.

[0143] The design supply and return water temperature difference refers to the difference between the supply and return water temperatures set during the design of the chilled water system. It serves as a baseline parameter for the flow coupling model and is used to limit the occurrence of low temperature difference syndrome. This is a system design parameter, obtained by consulting HVAC design specifications or project design drawings. A preferred value is five or eight degrees Celsius, and it must match the design cooling capacity, pipe diameter, and other parameters of the chilled water system.

[0144] The low temperature difference sensitivity coefficient is a parameter characterizing the degree to which the temperature difference between supply and return water decreases with increasing flow rate. Its dimension is the reciprocal of the flow rate (seconds per cubic meter), and it is used to quantify the coupling relationship between flow rate and temperature difference. It is obtained by fitting the flow rate-temperature difference curve from historical system operating data. The preferred value range is 50,000 to 200,000 seconds per cubic meter, with a typical value of 100,000 seconds per cubic meter. A larger value indicates that the temperature difference decreases more rapidly with increasing flow rate.

[0145] The smoothing scale parameter controls the degree of curvature of the smoothing activation function, ensuring the numerical stability of flow calculations and avoiding zero or negative denominators. It is a preset parameter, preferably ranging from 1.0 to 5.0, with a typical value of 2.0. Smaller values ​​result in more pronounced function curvature and stronger constraints on flow. The value must be verified through offline simulation to ensure smooth flow regulation and stable system operation.

[0146] The chiller unit's cooling load refers to the cooling power that the chiller unit should undertake within the next (t+k)th control time step. Its function is to compensate for the reduction in cold storage output in the middle stage, ensuring that the total cooling capacity of the system meets the terminal demand. It is calculated based on the principle of cooling capacity balance, that is, the predicted value of the system's required cooling capacity trajectory at that moment minus the target cold storage power after energy efficiency gating correction, in kilowatts. The value is largest in the middle stage of the cold storage charging and discharging state and smaller at both ends, which is complementary to the target cold storage power.

[0147] The pump operating frequency setpoint refers to the target operating frequency of the distribution pump within the next (t+k)-th control time step. It is the core command for adjusting the pump speed and controlling the flow rate. Calculated based on the pump similarity law, it is the ratio of the rated operating frequency of the distribution pump multiplied by the target flow rate setpoint at that moment to the rated flow rate of the chilled water system, expressed in Hertz. The value varies with the target flow rate, tending to be lower in the middle and higher at both ends to ensure the pump operates within its high-efficiency range.

[0148] The rated operating frequency of a water distribution pump refers to the standard operating frequency at the time of its design, and is the benchmark parameter for calculating the pump's operating frequency setting. This value can be obtained by consulting the pump's nameplate; it is usually the power grid frequency, i.e., 50 Hz or 60 Hz, and is a fixed value that does not require dynamic adjustment.

[0149] The rated flow rate of a chilled water system refers to the standard circulation flow rate designed for the system, serving as the benchmark parameter for calculating the pump operating frequency setting. It is obtained by consulting system design drawings or pump nameplates, and is expressed in cubic meters per second. This is a fixed value reflecting the flow rate requirement when the system is operating at full load.

[0150] like Figure 2 As shown, Figure 2 The system consists of a chiller unit, variable frequency distribution pumps, a chilled water network, terminal loads, and a eutectic salt packed bed cold storage tank, forming a chilled water circulation loop. The control process begins with the data acquisition module, which collects operating parameters from the chiller unit, pumps, cold storage tank, and terminal loads in real time via dotted lines. The collected data then enters the large model prediction module, which outputs predicted values ​​of the system's cooling demand and wet-bulb temperature for future moments. This prediction result is sent to the flow resistance sensitivity calculation module to quantify the hydraulic resistance characteristics of the packed bed, and then input to the energy efficiency gating module to execute the optimization algorithm. After gating correction, the system generates two control targets: flow rate setting and load setting. The flow rate setting is converted into a frequency command to directly control the speed of the distribution pumps, while the load setting is used to adjust the cooling output of the chiller unit. Finally, these commands are sent to the underlying devices via the PLC controller through Modbus or BACnet communication protocols, realizing energy efficiency-priority automated control of the entire cold storage central air conditioning system.

[0151] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A time-sharing optimization control method for a cold storage central air conditioning system based on a large model, characterized in that, include: Historical operating parameters and future environmental conditions of the chilled water system are collected and input into a pre-trained multivariate time series prediction model, which outputs the system's cooling demand trajectory and outdoor wet-bulb temperature change trajectory during the future prediction period. Based on the volumetric deformation characteristics of eutectic salt encapsulated capsules during phase change, a variable porosity flow resistance model is established to correlate the porosity of the packed bed with the cold storage and charging / discharging state. Based on the variable porosity flow resistance model, the flow resistance sensitivity coefficient, which characterizes the sensitivity of fluid resistance to flow rate changes, is calculated. Based on the outdoor wet-bulb temperature change trajectory, the energy efficiency ratio change trend of the chiller unit at different times is deduced to determine the cooling weight, and the cooling load allocation value is calculated in combination with the system cooling demand trajectory; An energy efficiency gating function is constructed using the flow resistance sensitivity coefficient to perform nonlinear correction on the cold storage load allocation value as the cold storage charging and discharging state changes. This generates a target flow setting trajectory where the flow rate decreases in the middle section and increases at both ends during the cold storage charging and discharging state. Based on this, the operating frequency of the distribution pump and the cooling load of the chiller are adjusted.

2. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 1, characterized in that, Historical operating parameters and future environmental operating parameters of the chilled water system, including: The historical operating parameters and future environmental condition parameters of the chilled water system are divided into static structural parameters, historically observable parameters, and known future environmental condition parameters. The static structural parameters include the capsule size of the eutectic salt encapsulation capsule, the bed length of the packed bed, the cross-sectional area of ​​the packed bed, and the initial porosity of the packed bed. The historically observable parameters include the historical supply water temperature, historical return water temperature, historical system flow rate, historical packed bed pressure difference, historical pump power, historical chiller unit power, and historical estimated values ​​of cold storage charge and discharge status. The known future environmental condition parameters include time-coded information for future times and predicted outdoor wet-bulb temperature for future times.

3. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 1, characterized in that, The pre-training process of a multivariate time series prediction model includes: A training sample set is constructed from the historical database of the chilled water system using a sliding time window approach. Each training sample contains an input feature vector and a label sequence, where the label sequence represents the true value of the system's cooling demand within the future prediction period. The quantile loss function is used as the training objective function for the multivariate time series prediction model. During training, the quantile regression loss between the predicted system cooling demand trajectory output by the multivariate time series prediction model at different quantiles and the label sequence is calculated. The model parameters of the multivariate time series prediction model are updated with the goal of minimizing the total loss of all training samples at all prediction steps and all quantiles.

4. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 1, characterized in that, The trajectory of system cooling demand and outdoor wet-bulb temperature changes during the forecast period include: The system's cooling demand trajectory and the outdoor wet-bulb temperature change trajectory are determined by selecting the predicted value corresponding to the median probability from the quantile prediction results output by the multivariate time series prediction model. The current cold storage charge / discharge state is calculated as follows: the instantaneous cold storage power is calculated based on the real-time supply and return water temperature difference and the real-time system flow rate of the chilled water system; the thermal front degradation factor is calculated based on the rate of change of the supply and return water temperature difference to characterize the loss of cold energy quality caused by thermal stratification degradation inside the packed bed; the product of the instantaneous cold storage power and the thermal front degradation factor is integrated over time to obtain the effective cold storage energy; and the effective cold storage energy is mapped to a cold storage charge / discharge state normalized between zero and one using a smoothing mapping function.

5. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 4, characterized in that, Based on the volumetric deformation characteristics of eutectic salt encapsulated capsules during phase transition, a variable porosity flow resistance model is established to correlate the porosity of the packed bed with the cold storage and discharging state, including: The porosity of the packed bed is defined as a quadratic function of the cold storage and charging / discharging state, and this quadratic function exhibits a concave shape in the middle section of the cold storage and charging / discharging state to simulate the porosity reduction effect caused by the volume expansion of the eutectic salt encapsulation capsule in the middle of the phase transition. Based on the Eugen equation, a nonlinear relationship between the packed bed pressure difference, the packed bed porosity, and the apparent flow velocity of the fluid is established, and a variable porosity flow resistance model is constructed.

6. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 5, characterized in that, The flow resistance sensitivity coefficient, which characterizes the sensitivity of fluid resistance to flow rate changes, is calculated based on the variable porosity flow resistance model. This includes: A flow resistance sensitivity coefficient is constructed using the porosity parameter in the variable porosity flow resistance model. This flow resistance sensitivity coefficient is equal to the difference between the current packed bed porosity and the current packed bed porosity, divided by the cube of the current packed bed porosity. The flow resistance sensitivity coefficient is used to quantify the amplification effect of the porosity term on system energy consumption in the nonlinear pressure drop characteristics described by the Eugen equation under the current cold storage charging and discharging state. The flow resistance sensitivity coefficient reaches its peak in the middle of the cold storage charging and discharging state, thereby identifying the high hydraulic resistance range.

7. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 1, characterized in that, Based on the outdoor wet-bulb temperature change trajectory, the energy efficiency ratio (EER) of the chiller unit at different times is deduced to determine the cooling load weight, and the cooling load allocation value is calculated in conjunction with the system's cooling demand trajectory, including: Based on the outdoor wet-bulb temperature change trajectory and system cooling demand trajectory at each future moment output by the multivariate time-series prediction model, the energy efficiency ratio (EER) of the chiller unit at each future moment is predicted using a linear regression model. An exponential function about the reciprocal of the chiller unit's EER is constructed to calculate the cooling release weight, such that the lower the chiller unit's EER, the greater the cooling release weight. The cooling release weights at all moments are then normalized. The total available cooling release capacity is calculated based on the current cold storage charging and discharging state and the preset minimum cold storage state. The total available cooling release capacity is then allocated to each moment of the future prediction period according to the cooling release weights, resulting in a cold storage load allocation value sequence.

8. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 6, characterized in that, Constructing an energy efficiency gating function using the aforementioned flow resistance sensitivity coefficient includes: An energy efficiency gating function is constructed, which is negatively correlated with the flow resistance sensitivity coefficient. This function is used to reduce the weight of the cold storage output in the range where the flow resistance sensitivity coefficient increases. The specific calculation logic of the energy efficiency gating function is as follows: take the value and divide it by the denominator, where the denominator is the sum of the product of the value, the pinch penalty strength parameter, and the flow resistance sensitivity coefficient. The pinch penalty strength parameter is used to adjust the degree of suppression in the high flow resistance range. When the cold storage charging / discharging state is in the middle stage, causing the flow resistance sensitivity coefficient to increase, the energy efficiency gating function is correspondingly reduced.

9. A time-sharing optimization control method for a cold storage central air conditioning system based on a large model, as described in claim 7 or 8, characterized in that, The cold storage load allocation value is nonlinearly corrected according to the cold storage charging and discharging state, generating a target flow setting trajectory with reduced flow in the middle section and increased flow at both ends during the cold storage charging and discharging state, including: The energy efficiency gating function is used to multiply and correct the cold storage load allocation value to obtain the target cold storage power sequence after hydraulic pinch suppression, thereby reducing the cold storage power in the middle of the cold storage charging and discharging state. Based on the low temperature difference syndrome characteristics of the chilled water system, a flow coupling model including the design supply and return water temperature difference and the low temperature difference sensitivity coefficient is introduced. The low temperature difference syndrome characteristic is that an increase in flow rate will lead to a decrease in the supply and return water temperature difference. A flow rate calculation formula is constructed using a smooth activation function to convert the target cold storage power sequence into a target flow rate setting trajectory. Under the same cooling demand, if the flow rate is too large and the temperature difference decreases, the increase in flow rate is limited by nonlinear constraints to prevent the system energy efficiency from deteriorating.

10. The time-sharing optimization control method for a cold storage central air conditioning system based on a large model according to claim 9, characterized in that, Adjusting the operating frequency of the water distribution pumps and the cooling load of the chiller unit, including: Based on the principle of cooling capacity balance, the target cold storage power sequence is subtracted from the system cooling demand trajectory output by the multivariate time-series prediction model to obtain the chiller unit's cooling load at each future time. This ensures that the chiller unit bears more load during the middle of the cold storage charging and discharging process to compensate for the reduction in cold storage output. Based on the pump similarity law, the pump operating frequency setpoints of the distribution pumps at each future time are calculated according to the ratio of the target flow rate setpoint trajectory to the rated flow rate of the chilled water system, combined with the rated operating frequency of the distribution pumps. The chiller unit cooling load and the pump operating frequency setpoints are then sent as time-series control commands to the underlying control system for execution.

Images